Epileptic Seizures Detection Using Visibility Graph Method and U-Net Architecture

In this study an innovative method for the epileptic seizures detection (ESD) have been introduced in which visibility graph (VG) techniques integrated with deep learning. The method transforms single-channel EEG signals as well as five sub-bands—delta, theta, alpha, beta, and gamma—to extract 16 features from the adjacency matrix of the VG. Then these features were given to a deep classifier utilizing U-Net architecture. An accuracy of 100% have been achieved in distinguishing of ictal (seizure) and non-ictal (seizure-free) signals, as well as in distinguishing normal from epileptic signals within the Bonn dataset. The proposed method also demonstrates an accuracy of 100% five subsets of the Bonn dataset, thereby exceeding the performance of existing research. Additionally, our method is capable of detecting seizures within a signal window of less than five seconds.